LAUSR

Review texts, which have been shown helpful for recommending items for users, are often available in the form of user feedback for items. Despite the success of previous approaches exploring reviews for recommendations, they are all based on long review texts. The users reviews are, however, often short in real-world applications. In this paper, we develop a novel approach to leverage information from short review texts for recommendation based on word vector representations. We first build word vectors to represent items and users, which are called item-vector and user-vector, respectively. After that we concatenate item-vectors and user-vectors to form a set of training data with the rating scores that users give to items. Finally we train a regression model to predict the unknown rating scores. In our experiment, we show that our approach is effective, compared to state-of-the-art algorithms.